Skip to content

Improved detection of seismicity at Cordón Caulle (2011-12) using multiple machine learning approaches

Diego Lobos-Lillo (1); Diana Roman (2); Matt Pritchard (1); Carlos Cardona (3) ; Luis Franco (3); Francisco Delgado (4)

  • Affiliations: (1) Earth and Atmospheric Science, Cornell University, Ithaca, NY, USA; (2) Carnegie Institution for Science, Washington D.C., USA; (3) Observatorio Volcanológico de los Andes del Sur, SERNAGEOMIN, Temuco, Chile; (4) Departamento de Ingenieria, Universidad de Chile, Santiago, Chile.

  • Presentation type: Poster

  • Presentation time: Thursday 16:30 - 18:30, Room Poster Hall

  • Poster Board Number: 115

  • Programme No: 2.1.25

  • Theme 2 > Session 1

Abstract

The 2011-12 Cordón Caulle, Chile, eruption provides the best case study of seismic data spanning a silicic eruption case study for understanding silicic volcanic system dynamics. Despite significant insights from seismic data, questions remain regarding eruptive dynamics, magma intrusions, precursory seismicity patterns, and tectonic-magmatic interactions. This study applies machine learning techniques for seismic event detection (VT and LP) and location. Our workflow integrates three deep learning phase-pickers (GPD, PhaseNet, EQTransformer) and dual phase associators (GaMMA, PyOcto), combining probabilistic detection with nonlinear location methods to reveal refined spatiotemporal patterns throughout the eruption sequence. Analysis of over 6,000 relocated events shows systematic variations in hypocenter distributions across eruptive phases, extending the recognized unrest duration from 5 to 8 months and tripling event detection compared with manual catalogs. Enhanced detection reveals peak rates of 900 events per day compared with 140 in conventional catalogs. We identify complex precursory swarm sequences and spatial correlations between seismicity and InSAR-derived deformation sources. This refined analysis, integrated with geodetic observations, provides new perspectives on fluid dynamics on the hydrothermal system, magmatic processes and volcano-tectonic interactions during silicic eruptions. Our results demonstrate machine learning's value in extracting additional insights from monitoring data, with implications for volcanic hazard assessment